Table 2.

Classification results from multilevel PLS-DA and PLS-DA

				Class. model	No. of LVs	Variance X/Y (%)	Sensitivity/ Specificity (%)	Class. accuracy (%)	Permutation p-value
Paired data	All samples	Control vs Treatment	(n = 65p)	Multilevel PLS-DA	2	58.9/50.9	87.9/87.9	87.9	< 0.001
	Partial response	Control vs Treatment	(n = 37p)	Multilevel PLS-DA	2	55.7/61.7	80.0/80.0	80.0	< 0.001
	Stable disease	Control vs Treatment	(n = 28p)	Multilevel PLS-DA	2	63.2/61.3	88.7/88.7	88.7	< 0.001
	Survivors	Control vs Treatment	(n = 44p)	Multilevel PLS-DA	2	60.0/50.5	83.0/83.0	83.0	< 0.001
	Non-survivors	Control vs Treatment	(n = 19p)	Multilevel PLS-DA	2	70.6/63.9	82.5/82.5	82.5	0.006
Unpaired data	All samples	Pre- vs Post-treatment	(n = 65p)	PLD-DA	2	50.3/29.2	80.7/57.1	68.9	< 0.001
	Pre-treatment	Partial response vs Stable disease	(n = 48) (n = 32)	PLS-DA	2	51.5/19.3	56.7/60.2	58.4	0.231
	Post-treatment	Partial response vs Stable disease	(n = 41) (n = 29)	PLS-DA	NaN	-	-	-	-
	Pre-treatment	Survivors vs Non-survivors	(n = 57) (n = 20)	PLS-DA	NaN	-	-	-	-
	Post-treatment	Survivors vs Non-survivors	(n = 47) (n = 21)	PLS-DA	3	62.8/32.1	58.4/75.3	70.1	0.009
	Post-treatment	Epirubicin vs Paclitaxel	(n = 29) (n = 23)	PLS-DA	2	45.7/26.8	45.0/48.3	46.5	0.245

The sensitivity is for detecting a treatment/stable disease/non-survivor/Paclitaxel spectrum; Variance X/Y, amount of variance from X/Y explained by the model; NaN, no valid model; p, pairs